intro

To illustrate the regression case, we’ll use the Hitters data from the ISLR2 package. A sample of the data are shown below.

data("Hitters", package = "ISLR2")

# Remove rows with missing response values
head(hitters <- Hitters[!is.na(Hitters$Salary), ])
#>                   AtBat Hits HmRun Runs RBI Walks Years CAtBat CHits CHmRun
#> -Alan Ashby         315   81     7   24  38    39    14   3449   835     69
#> -Alvin Davis        479  130    18   66  72    76     3   1624   457     63
#> -Andre Dawson       496  141    20   65  78    37    11   5628  1575    225
#> -Andres Galarraga   321   87    10   39  42    30     2    396   101     12
#> -Alfredo Griffin    594  169     4   74  51    35    11   4408  1133     19
#> -Al Newman          185   37     1   23   8    21     2    214    42      1
#>                   CRuns CRBI CWalks League Division PutOuts Assists Errors
#> -Alan Ashby         321  414    375      N        W     632      43     10
#> -Alvin Davis        224  266    263      A        W     880      82     14
#> -Andre Dawson       828  838    354      N        E     200      11      3
#> -Andres Galarraga    48   46     33      N        E     805      40      4
#> -Alfredo Griffin    501  336    194      A        W     282     421     25
#> -Al Newman           30    9     24      N        E      76     127      7
#>                   Salary NewLeague
#> -Alan Ashby        475.0         N
#> -Alvin Davis       480.0         A
#> -Andre Dawson      500.0         N
#> -Andres Galarraga   91.5         N
#> -Alfredo Griffin   750.0         A
#> -Al Newman          70.0         A

We’ll start by fitting a basic EBMRegressor to the hitters data set. Note that the ebm() function currently only supports the usual formula interface.

library(ebm)

# Fit a default EBM regressor
fit <- ebm(Salary ~ ., data = hitters, objective = "rmse")
fit  # still need to implement print() and summary() methods
#> ExplainableBoostingRegressor(early_stopping_tolerance=0)

You can obtain predictions using the familiar predict() method employed by most modeling packages in R. Note that thorugh bagging, EBMs can provide standard errors for the predictions if requested.

head(predict(fit, newdata = hitters))
#> [1] 489.4548 626.1970 870.3430 169.8797 659.6543 270.5382
head(predict(fit, newdata = hitters, se.fit = TRUE))
#>          [,1]      [,2]
#> [1,] 489.4548  53.92471
#> [2,] 626.1970  98.39167
#> [3,] 870.3430 241.04199
#> [4,] 169.8797  43.14385
#> [5,] 659.6543  54.02381
#> [6,] 270.5382 155.71903

You can produce several plotly-based graphs to help interpret the output of "EBM" objects using the generic plot() method; this function supports both global and local interpretations. The default simply prints a global measure of importance based on the sum of the absolute value of each variable sterm contributions. (For Markdown-type documents, like this vignette, you need to specify display = "markdown"; see ?ebm::plot for details.)

plot(fit, display = "markdown")

You can also plot the individual shape functions (or term contributions), as shown below:

plot(fit, term = "Years", display = "markdown")

While the ebm package does not expose 100% of the functionality available in Python, you can pretty much do anything you need by interacting directly with the underlying Python objects (the magic happens through reticulate). For instance, we invoke the monottonize() method to post process the term for "Years" by enforcing increasing monotonicity:

fit$monotonize("Years", increasing = FALSE)
#> ExplainableBoostingRegressor(early_stopping_tolerance=0)
plot(fit, term = "Years", display = "markdown")

You can also display local explanations (though, one at a time) by specifying local = TRUE:

# Understand an individual prediction
x <- subset(hitters, select = -Salary)[1L, ]  # use first observation
plot(fit, local = TRUE, X = x, y = hitters$Salary[1L], display = "markdown")